Review Session

Analyze this session's tool and skill usage. Evaluate success and provide improvement recommendations.

Step 1: Gather Session Context

Review the full conversation history and identify:

1. •Tools used - All invocations (Bash, Read, Write, Edit, Grep, Glob, Task, WebFetch, etc.)
2. •Skills invoked - Any /skill-name skills used
3. •Agent delegations - Task tool calls to subagents
4. •User questions asked - AskUserQuestion tool usage

For each usage, note:

• •Intent
• •Outcome (success, failure, partial)
• •Number of attempts needed
• •Any errors or retries

Step 2: Identify Patterns

Analyze for these pattern types:

User Corrections (PRIMARY SIGNAL):

• •User said "no, I meant..." or "actually..."
• •User rephrased or repeated a request
• •User provided additional context after seeing initial results
• •User pointed out something missed
• •User modified or adjusted output/approach
• •Any indication first attempt did not meet expectations

Inefficiencies:

• •Multiple reads of same file that could have been batched
• •Searches too broad or too narrow
• •Repeated similar searches indicating unclear requirements
• •Bash used where specialized tools better (e.g., cat instead of Read)

Errors:

• •Tool failures and root causes
• •Retry loops indicating misunderstanding
• •Permission or sandbox issues
• •Incorrect file paths or patterns

Successes:

• •Effective agent delegation for complex tasks
• •Efficient search strategies
• •Appropriate tool selection
• •Requests completed without user correction

Step 3: Evaluate Each Major Task

For each task or request, assess:

Step 4: Generate Recommendations

Prepare recommendations in these categories:

• •Immediate Improvements - Better tool choices, search strategies, how to anticipate user corrections
• •Process Improvements - Patterns to adopt, effective tool combinations, when to delegate
• •Learning Points - New capabilities discovered, edge cases encountered, configuration adjustments

Step 5: Present Report

Structure the report:

Session Review Summary

• •Primary tasks addressed
• •Overall success assessment

Tool Usage Analysis

• •Table: Tool | Count | Success Rate | Notes
• •Efficiency score: Excellent / Good / Moderate / Needs Improvement

User Corrections Required For each correction:

• •What was attempted
• •What user had to clarify or correct
• •How this could have been anticipated

Other Issues For each issue:

• •What happened
• •Impact on session
• •Recommended alternative

Successful Patterns For each pattern:

• •Why it worked well
• •When to use again

Recommendations

• •For future sessions
• •Tool selection guidance
• •Configuration suggestions

Step 6: Offer Follow-up

Ask the user if they want to:

1. •Deep dive on a specific area
2. •Export recommendations to .claude/session-reviews/ with timestamped filename
3. •No further action